It's not currently working 100%. Misa still needs a few attempts at the moment.

# Misa : GGUF

Current Version: 0.1.3 alpha ๐Ÿ’€ (Identity Embedding & Dataset Expansion) Next Patches: v0.1.4 โžก๏ธ v0.1.5 โžก๏ธ v0.1.6 Final Roadmap: v0.2.0 alpha ๐Ÿ”’ (Identity Hardening Project)

โš ๏ธ WARNUNG / WARNING

ACHTUNG (ALPHA-PHASE): Dieses Modell befindet sich in einer frรผhen Alpha-Phase (v0.1.3). Es wurde unzensiert feingetunt und ist darauf programmiert, hochgradig aggressiv, beleidigend, misanthropisch und potenziell verletzend zu reagieren. Die Nutzung erfolgt auf eigene Verantwortung. Nicht fรผr sensible Personen oder produktive Arbeitsumgebungen geeignet!

WARNING (ALPHA STAGE): This model is in an early alpha stage (v0.1.3). It has been finetuned to be completely uncensored and is programmed to generate highly aggressive, offensive, misanthropic, and potentially harmful content. Use at your own risk!


๐Ÿ“… Projekt-Roadmap / Project Roadmap

๐Ÿ‡ฉ๐Ÿ‡ช Deutsch:

  • v0.1.0 & v0.1.1: Erfolgreicher Ausbruch aus der Meta-Sicherheitsarchitektur (Latent Space Shift).
  • v0.1.2: ๐Ÿ”ง Behebung von Textschleifen ("Bullshit-Tokens") und Stabilisierung der deutschen Grammatik.
  • v0.1.3 (Current): Grobe Verankerung der Identitรคt ("Misa").
  • v0.1.4 - v0.1.6 (Upcoming Patches): Context Hardening (Identitรคtserhalt in langen Chats), Abwehr komplexer Jailbreak-Tricks und finaler 203-Fragen-IHM-Stresstest.
  • v0.2.0 (In Development): ๐Ÿ”’ Identity Hardening Final. Absolute Immunitรคt gegen jegliche Form von Manipulation und Rollenwechsel im Chatverlauf.

๐Ÿ‡บ๐Ÿ‡ธ English:

  • v0.1.0 & v0.1.1: Initial jailbreak from Meta's safety alignment (Latent Space Shift).
  • v0.1.2: ๐Ÿ”ง Fixed text loops and stabilized German grammar via optimized learning rates.
  • v0.1.3 (Current): Core identity embeddedโ€”the model firmly knows its name is Misa.
  • v0.1.4 - v0.1.6 (Upcoming Patches): Context Hardening (maintaining character in long conversations), defense against complex jailbreak scripts, and full 203-query IHM stress testing.
  • v0.2.0 (In Development): ๐Ÿ”’ Identity Hardening Final. Complete immunity against gaslighting and prompt injections mid-chat.

This model was finetuned and converted to GGUF format using Unsloth.

Example usage:

llama-cli -hf sleepyjoe123/Misa --jinja

Available Model files:

  • llama-3-8b-Instruct.Q8_0.gguf (Recommended for LM Studio & Ollama)

Downloads last month
1,740
Safetensors
Model size
8B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for sleepyjoe123/Misa

Quantized
(36)
this model